Stability of the AHAAH Model
This information paper is written to all who are concerned with the performance and stability of the AHAAH model. It has been brought to our attention there may be a misunderstanding within the technical community concerning AHAAH. We understand several parties have complained that AHAAH is “a moving target,” being adjusted each time a problem is found with the model. At two levels, that is simply not true.
First, the core programming – that dealing with the propagation of sound into the ear and the calculation of risk – has been unchanged since the model was publicly released in the mid -1990s.
Secondly, it has always been our expectation that if data are produced showing that the model’s predictions are in error, it would be in keeping with the best principles of science and model-making to determine which properties of the ear have been incorrectly represented in the model and to change them so that the model’s performance matches all available data as well as possible. We await the arrival of such data.
The issue regarding adjustments to AHAAH has, no doubt, come about because there have been numerous changes made to the overall AHAAH software package to introduce new features to:
- Include hearing protection
- Improve how the program displays calculated results
- Improve the code so that it does not become unstable under unusual circumstances
The most prominent changes have been associated with the hearing protection module (HPM). The AHAAH model requires a waveform for input, in the free field (if no protection is worn), at the ear canal entrance, or at the eardrum location if hearing protection is worn. A measured waveform on an acoustical test fixture (ATF; manikin) or human subjects might be used. Lacking human or ATF data for the AHAAH model to process, some method of calculating the waveform under the hearing protection device (HPD) was required. Initially, a minimum phase approach was used to calculate the waveform for processing (attenuations “typical” of single and double HPDs were used – a considerable improvement over the single number adjustment for hearing protection found in MIL-STD 1747D).
More recently, a full electro-acoustic HPD model was developed which allows calculations with specific types of hearing protectors, including those that are non-linear. In summary, the hearing protection module has been adjusted many times to better represent the full range of hearing protector effects, but the results produced by the core AHAAH model have not changed since it was originally released.
A few years ago, a feature was added which allows the calculation of hazard for any percentile of susceptibility (the basic concept is inherent in the model). While the U.S. Army might choose to protect the 95%ile ear, other agencies or governments might choose some other level of protection or scientists may wish to calculate effects on other populations and they are free to do so.
There have also been minor changes in the computational algorithms used in AHAAH to correct computational problems as they were discovered. We found that calculations with double hearing protection caused the waveform under the combined protector to be much longer than the initial waveform (a reasonable outcome). In order to avoid the error that would result (the longer waveform would “wrap around” and appeared to be additional information at the beginning of the waveform), the end of the waveform should be padded with “zeros.” Recent versions of the AHAAH model do this automatically (previously, it was a manual process). If in doubt, append zeros to the end of the waveform. (It is not unreasonable for 75% of the waveform to be zeros). The acuity of the AHAAH model has improved as data acquisition equipment advanced and equipment costs were reduced. Currently, the recommended sampling rate for acquiring the waveform is 192 kHz (minimum). This sampling rate is adequate to avoid errors that might result from under-sampling. As recently as 10-years ago, data were recorded at sampling rates as low as 44 kHz.
In spite of these changes, we still are using the same basic algorithms. However, the fix for “wrap-around” described above did produce an unintended consequence. Namely, the numerical integrations in the outer and middle ear use the fourth order Runga-Kutta method with a step size control based on the Richardson criterion. (i.e.; calculation over a time interval is repeated with half the step size and the results are compared.) This works well unless the waveform ends with appended zeros, in which case the procedure stops. To prevent this, we added dithering noise to the incoming waveform at a level below the sampling resolution. This noise causes slight changes (in the fourth significant digit) in the calculated results from run to run.
Finally, we note that some features of the AHAAH model might result in “surprises” in the data analysis and make it appear that the model had changed. The instructions for the use of the model cover these issues; but experience shows users may make the common mistakes described below.
- Establish start: The “Establish Start” function on the right-click pop-up menu allows the user to synchronize the aural reflex onset with a point on the waveform. If this is not done consistently, then the un-warned hazard can change while the warned hazard does not change.
- Measurement location: The waveform to be analyzed should be measured at one of three locations designated by a “1” for free-field, “2” for ear canal entrance and “3” for eardrum (or ATF microphone). When a waveform is first imported, this location number in the header of the “.AHA” file defaults to “1” (assumes free field recording). If another location is appropriate, the user needs to change the digit to reflect the appropriate location and save the waveform after editing. If this is not done, the waveform might be applied at the wrong location, generating an erroneous result.
- Waveform importation: During the importation of a waveform, the default sampling rate is set at 200 kHz. If the imported waveform is a “.WAV” file, then AHAAH reads the sampling frequency from the WAV file header. Otherwise, the default sampling rate may not be correct and that will cause erroneous results.
- Taper ends: If the waveform is not tapered at the ends, then end-effect distortions can occur generating an erroneous result.
Finally, we note that AHAAH began as a research tool and was designed to be as flexible as possible so that any researcher might test its provisions and perhaps develop a better model. With that in mind, all of the model’s variables were made accessible in the “man.coe” file. That file also contains “features” that are not commonly used; but which can be turned off or on at will. (e.g.; it is possible to print all the hazard values calculated, not just the highest, which is what is commonly reported.) In the world of damage risk criteria, however, too many options are the enemy of a standard. Hence, the version of AHAAH intended for standardization has been structured to minimize choices that might result in an inaccurate calculation. So, if the “man.coe” file is modified from its original form AHAAH will indeed be a moving target (but of your own making). In that regard, it might be useful to have a copy of the original “man.coe” file stored as “adam.coe” – the original man.
In conclusion, we note that even though features have been added to the AHAAH model, the basic calculation of hazard has not been changed. Nonetheless, because the model is structured to be conformal with the ear’s anatomy; if new data are produced, the model is in a form that promotes improvement. We invite it!